Skip to content

hackathon_demo_script.md

Latest commit

 

History

History
83 lines (70 loc) · 9.25 KB

hackathon_demo_script.md

File metadata and controls

83 lines (70 loc) · 9.25 KB
Raw

Sharibo — demo script v2 (~2:20)

Why v2: the hackathon asks for a 2–3 minute walkthrough. v1 was 60s — punchy, but it spent zero seconds on the project's strongest differentiator: real on-chain Groth16 verification, with a from-the-trenches engineering story (the BN254 wall) and public tx hashes. v2 keeps v1's pacing for the first minute, then adds the two sections judges actually score on: "why this was hard" and "what we verified vs. what we're honest about."

Tone: same as v1 — fast, confident. Every claim below is backed by full_product_breakdown.md §4/§6/§13. Nothing needs softening; it's all real.

Shot-by-shot

Time On screen Voiceover
0:00–0:04 Landing. Name-wall chips visible. One beat of stillness. "Every culture on Earth runs a version of this."
0:04–0:10 Pan across chips: ajo, tanda, susu, tontine, hui, chit fund… "Ajo. Tanda. Susu. Tontine. Hundreds of millions of people save in rotating circles — and every one runs on the same weak point: everyone knows exactly who's collecting the pot."
0:10–0:14 Cut to Sharibo. Click Launch a 5-member circle on testnet. "Sharibo fixes that. On-chain. For real — and I mean cryptographically real, which I'll prove in a minute."
0:14–0:22 Sped-up montage (2–3×): circle created → 5 funds → pot bar fills → ring nodes light. "Five members. Five real deposits on Stellar testnet — all public, all visible. That part isn't private, on purpose."
0:22–0:26 Select claimant → click Generate proof & claim. Groth16 · BLS12-381 · 1,452 constraints line holds on screen. "Now — who gets the pot? Watch."
0:26–0:36 Proving runs in REAL TIME (never speed this up). Then claim success → ring reveal → outside "?" node. "That's not a spinner. That's a Groth16 proof being generated live in the browser — no server, no trusted party — and verified on-chain by a real BLS12-381 pairing check inside a Soroban contract."
0:36–0:42 Ring caption / explorer: 5 deposits, 1 payout, fresh address. Overlay: 5 in → 1 out → 0 links. "Five deposits in. One payout out — to a brand-new address that's never touched this circle. Nobody, including me, can tell you which member just got paid."
0:42–0:50 Click Try to claim again. Red rejection box, full-bleed. Overlay: Error(Contract, #4): AlreadyClaimed. "Now watch someone try to cheat — replaying that exact proof. Rejected. On-chain. That's not a UI check — the contract's nullifier map refused it."
0:50–1:00 Terminal or slide: tampered-proof rejection — Error(Contract, #5): InvalidProof. "And a tampered proof? Also rejected — the pairing equation itself fails. Both rejections are reproducible from the repo."
— SECTION: WHY THIS WAS HARD
1:00–1:12 Simple graphic or code view. Overlay: BN254 pairing ≈ 560,000,000 instructions vs Soroban budget: 100,000,000. "Here's the part that separates this from a tutorial. The standard Circom stack uses the BN254 curve. We measured it: one BN254 pairing on Soroban costs about 560 million instructions — against a hard 100-million cap. Not expensive. Impossible."
1:12–1:24 Show the verifier code (pairing_check lines) or contract snippet. Overlay: claim(): 48.0M / 100M CPU (~48%). "So Sharibo runs the entire pipeline — circuit, trusted setup, contract — on BLS12-381, the one curve Stellar accelerates natively. A real claim, with a real proof, costs 48 million instructions. Comfortably inside budget. Measured, not estimated."
1:24–1:36 Show the Poseidon note briefly (one line), then cut to the tx-hash table from the README. "That meant sourcing Poseidon constants generated for the right field — cross-checked against Soroban's own modulus — and matching byte layouts across circuit, contract, and client. The accepted proof, the rejected proof, the full round: every one is a public testnet transaction. The hashes are in the README. Check them."
— SECTION: PROOF & HONESTY
1:36–1:48 Fast cut: circuit tests 5/5 green → contract tests 8/8 green → e2e output tail. "Five circuit tests. Eight contract tests — including the pairing check failing for a forged input, and a CPU-budget assertion. One end-to-end script that runs the whole round against live testnet, then replays the nullifier and proves it gets rejected."
1:48–2:00 Plain card: "Honest scope: claim-side privacy. One round. Testnet." "And the honest scope, stated plainly: privacy covers who claims, not who funds. One round is demoed, not a full rotation. It's testnet, with a single-party trusted setup. Every limitation is documented — nothing is silently faked."
2:00–2:12 Back to name-wall / logo. "Ajo, tanda, susu, tontine — this is how a huge share of the world already saves. Stellar is where real-world money already moves. Sharibo makes the trust those circles run on cryptographic instead of social."
2:12–2:20 Final card: SHARIBO + "Real ZK. Real chain. Real privacy." + repo URL + contract ID. "Sharibo. Real ZK, verified on a real chain — for a problem that's already real. The repo has everything: code, tests, transaction hashes. See for yourself."

Clean VO-only script

Every culture on Earth runs a version of this. Ajo. Tanda. Susu. Tontine. Hundreds of millions of people save in rotating circles — and every one runs on the same weak point: everyone knows exactly who's collecting the pot.

Sharibo fixes that. On-chain. For real — and I mean cryptographically real, which I'll prove in a minute.

Five members. Five real deposits on Stellar testnet — all public, all visible. That part isn't private, on purpose.

Now — who gets the pot? Watch.

That's not a spinner. That's a Groth16 proof being generated live in the browser — no server, no trusted party — and verified on-chain by a real BLS12-381 pairing check inside a Soroban contract.

Five deposits in. One payout out — to a brand-new address that's never touched this circle. Nobody, including me, can tell you which member just got paid.

Now watch someone try to cheat — replaying that exact proof. Rejected. On-chain. That's not a UI check — the contract's nullifier map refused it. And a tampered proof? Also rejected — the pairing equation itself fails. Both rejections are reproducible from the repo.

Here's the part that separates this from a tutorial. The standard Circom stack uses the BN254 curve. We measured it: one BN254 pairing on Soroban costs about 560 million instructions — against a hard 100-million cap. Not expensive. Impossible.

So Sharibo runs the entire pipeline — circuit, trusted setup, contract — on BLS12-381, the one curve Stellar accelerates natively. A real claim, with a real proof, costs 48 million instructions. Comfortably inside budget. Measured, not estimated.

That meant sourcing Poseidon constants generated for the right field — cross-checked against Soroban's own modulus — and matching byte layouts across circuit, contract, and client. The accepted proof, the rejected proof, the full round: every one is a public testnet transaction. The hashes are in the README. Check them.

Five circuit tests. Eight contract tests — including the pairing check failing for a forged input, and a CPU-budget assertion. One end-to-end script that runs the whole round against live testnet, then replays the nullifier and proves it gets rejected.

And the honest scope, stated plainly: privacy covers who claims, not who funds. One round is demoed, not a full rotation. It's testnet, with a single-party trusted setup. Every limitation is documented — nothing is silently faked.

Ajo, tanda, susu, tontine — this is how a huge share of the world already saves. Stellar is where real-world money already moves. Sharibo makes the trust those circles run on cryptographic instead of social.

Sharibo. Real ZK, verified on a real chain — for a problem that's already real. The repo has everything: code, tests, transaction hashes. See for yourself.

Overlays (do these — they carry credibility silently)

  • 0:22 — Groth16 · BLS12-381 · 1,452 constraints
  • 0:36 — 5 in → 1 out → 0 links
  • 0:42 — Error(Contract, #4): AlreadyClaimed
  • 0:50 — Error(Contract, #5): InvalidProof
  • 1:00 — BN254 pairing: ~560M instr · budget: 100M
  • 1:12 — real claim(): 48.0M / 100M CPU
  • 1:24 — tx: 2258…8087 · ledger 3379702
  • 2:12 — repo URL + CB64IZIB…2LCF

Recording checklist (unchanged items from v1 still apply)

  • FIRST: full click-through in a real browser. Still the only unverified layer. If the ring/stepper/proving flow breaks, fix before recording anything.
  • Pre-warm testnet off-camera (first RPC calls are the slow ones).
  • Capture the tests-passing footage (npm test in circuits/, cargo test, tail of npm run e2e) — three short terminal clips, 3–4s each.
  • Never speed up the proving wait; speed up the funding montage instead.
  • If cutting for time: trim the funding montage and §"Proof & honesty" test clips first. Never cut the proving moment, the two rejections, or the BN254→BLS12-381 story.